xref: /linux/fs/bcachefs/bcachefs_format.h (revision 3f41368fbfe1b3d5922d317fe1a0a0cab6846802)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_FORMAT_H
3 #define _BCACHEFS_FORMAT_H
4 
5 /*
6  * bcachefs on disk data structures
7  *
8  * OVERVIEW:
9  *
10  * There are three main types of on disk data structures in bcachefs (this is
11  * reduced from 5 in bcache)
12  *
13  *  - superblock
14  *  - journal
15  *  - btree
16  *
17  * The btree is the primary structure; most metadata exists as keys in the
18  * various btrees. There are only a small number of btrees, they're not
19  * sharded - we have one btree for extents, another for inodes, et cetera.
20  *
21  * SUPERBLOCK:
22  *
23  * The superblock contains the location of the journal, the list of devices in
24  * the filesystem, and in general any metadata we need in order to decide
25  * whether we can start a filesystem or prior to reading the journal/btree
26  * roots.
27  *
28  * The superblock is extensible, and most of the contents of the superblock are
29  * in variable length, type tagged fields; see struct bch_sb_field.
30  *
31  * Backup superblocks do not reside in a fixed location; also, superblocks do
32  * not have a fixed size. To locate backup superblocks we have struct
33  * bch_sb_layout; we store a copy of this inside every superblock, and also
34  * before the first superblock.
35  *
36  * JOURNAL:
37  *
38  * The journal primarily records btree updates in the order they occurred;
39  * journal replay consists of just iterating over all the keys in the open
40  * journal entries and re-inserting them into the btrees.
41  *
42  * The journal also contains entry types for the btree roots, and blacklisted
43  * journal sequence numbers (see journal_seq_blacklist.c).
44  *
45  * BTREE:
46  *
47  * bcachefs btrees are copy on write b+ trees, where nodes are big (typically
48  * 128k-256k) and log structured. We use struct btree_node for writing the first
49  * entry in a given node (offset 0), and struct btree_node_entry for all
50  * subsequent writes.
51  *
52  * After the header, btree node entries contain a list of keys in sorted order.
53  * Values are stored inline with the keys; since values are variable length (and
54  * keys effectively are variable length too, due to packing) we can't do random
55  * access without building up additional in memory tables in the btree node read
56  * path.
57  *
58  * BTREE KEYS (struct bkey):
59  *
60  * The various btrees share a common format for the key - so as to avoid
61  * switching in fastpath lookup/comparison code - but define their own
62  * structures for the key values.
63  *
64  * The size of a key/value pair is stored as a u8 in units of u64s, so the max
65  * size is just under 2k. The common part also contains a type tag for the
66  * value, and a format field indicating whether the key is packed or not (and
67  * also meant to allow adding new key fields in the future, if desired).
68  *
69  * bkeys, when stored within a btree node, may also be packed. In that case, the
70  * bkey_format in that node is used to unpack it. Packed bkeys mean that we can
71  * be generous with field sizes in the common part of the key format (64 bit
72  * inode number, 64 bit offset, 96 bit version field, etc.) for negligible cost.
73  */
74 
75 #include <asm/types.h>
76 #include <asm/byteorder.h>
77 #include <linux/kernel.h>
78 #include <linux/uuid.h>
79 #include <uapi/linux/magic.h>
80 #include "vstructs.h"
81 
82 #ifdef __KERNEL__
83 typedef uuid_t __uuid_t;
84 #endif
85 
86 #define BITMASK(name, type, field, offset, end)				\
87 static const __maybe_unused unsigned	name##_OFFSET = offset;		\
88 static const __maybe_unused unsigned	name##_BITS = (end - offset);	\
89 									\
90 static inline __u64 name(const type *k)					\
91 {									\
92 	return (k->field >> offset) & ~(~0ULL << (end - offset));	\
93 }									\
94 									\
95 static inline void SET_##name(type *k, __u64 v)				\
96 {									\
97 	k->field &= ~(~(~0ULL << (end - offset)) << offset);		\
98 	k->field |= (v & ~(~0ULL << (end - offset))) << offset;		\
99 }
100 
101 #define LE_BITMASK(_bits, name, type, field, offset, end)		\
102 static const __maybe_unused unsigned	name##_OFFSET = offset;		\
103 static const __maybe_unused unsigned	name##_BITS = (end - offset);	\
104 static const __maybe_unused __u##_bits	name##_MAX = (1ULL << (end - offset)) - 1;\
105 									\
106 static inline __u64 name(const type *k)					\
107 {									\
108 	return (__le##_bits##_to_cpu(k->field) >> offset) &		\
109 		~(~0ULL << (end - offset));				\
110 }									\
111 									\
112 static inline void SET_##name(type *k, __u64 v)				\
113 {									\
114 	__u##_bits new = __le##_bits##_to_cpu(k->field);		\
115 									\
116 	new &= ~(~(~0ULL << (end - offset)) << offset);			\
117 	new |= (v & ~(~0ULL << (end - offset))) << offset;		\
118 	k->field = __cpu_to_le##_bits(new);				\
119 }
120 
121 #define LE16_BITMASK(n, t, f, o, e)	LE_BITMASK(16, n, t, f, o, e)
122 #define LE32_BITMASK(n, t, f, o, e)	LE_BITMASK(32, n, t, f, o, e)
123 #define LE64_BITMASK(n, t, f, o, e)	LE_BITMASK(64, n, t, f, o, e)
124 
125 struct bkey_format {
126 	__u8		key_u64s;
127 	__u8		nr_fields;
128 	/* One unused slot for now: */
129 	__u8		bits_per_field[6];
130 	__le64		field_offset[6];
131 };
132 
133 /* Btree keys - all units are in sectors */
134 
135 struct bpos {
136 	/*
137 	 * Word order matches machine byte order - btree code treats a bpos as a
138 	 * single large integer, for search/comparison purposes
139 	 *
140 	 * Note that wherever a bpos is embedded in another on disk data
141 	 * structure, it has to be byte swabbed when reading in metadata that
142 	 * wasn't written in native endian order:
143 	 */
144 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
145 	__u32		snapshot;
146 	__u64		offset;
147 	__u64		inode;
148 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
149 	__u64		inode;
150 	__u64		offset;		/* Points to end of extent - sectors */
151 	__u32		snapshot;
152 #else
153 #error edit for your odd byteorder.
154 #endif
155 } __packed
156 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
157 __aligned(4)
158 #endif
159 ;
160 
161 #define KEY_INODE_MAX			((__u64)~0ULL)
162 #define KEY_OFFSET_MAX			((__u64)~0ULL)
163 #define KEY_SNAPSHOT_MAX		((__u32)~0U)
164 #define KEY_SIZE_MAX			((__u32)~0U)
165 
166 static inline struct bpos SPOS(__u64 inode, __u64 offset, __u32 snapshot)
167 {
168 	return (struct bpos) {
169 		.inode		= inode,
170 		.offset		= offset,
171 		.snapshot	= snapshot,
172 	};
173 }
174 
175 #define POS_MIN				SPOS(0, 0, 0)
176 #define POS_MAX				SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, 0)
177 #define SPOS_MAX			SPOS(KEY_INODE_MAX, KEY_OFFSET_MAX, KEY_SNAPSHOT_MAX)
178 #define POS(_inode, _offset)		SPOS(_inode, _offset, 0)
179 
180 /* Empty placeholder struct, for container_of() */
181 struct bch_val {
182 	__u64		__nothing[0];
183 };
184 
185 struct bversion {
186 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
187 	__u64		lo;
188 	__u32		hi;
189 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
190 	__u32		hi;
191 	__u64		lo;
192 #endif
193 } __packed
194 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
195 __aligned(4)
196 #endif
197 ;
198 
199 struct bkey {
200 	/* Size of combined key and value, in u64s */
201 	__u8		u64s;
202 
203 	/* Format of key (0 for format local to btree node) */
204 #if defined(__LITTLE_ENDIAN_BITFIELD)
205 	__u8		format:7,
206 			needs_whiteout:1;
207 #elif defined (__BIG_ENDIAN_BITFIELD)
208 	__u8		needs_whiteout:1,
209 			format:7;
210 #else
211 #error edit for your odd byteorder.
212 #endif
213 
214 	/* Type of the value */
215 	__u8		type;
216 
217 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
218 	__u8		pad[1];
219 
220 	struct bversion	version;
221 	__u32		size;		/* extent size, in sectors */
222 	struct bpos	p;
223 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
224 	struct bpos	p;
225 	__u32		size;		/* extent size, in sectors */
226 	struct bversion	version;
227 
228 	__u8		pad[1];
229 #endif
230 } __packed
231 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
232 /*
233  * The big-endian version of bkey can't be compiled by rustc with the "aligned"
234  * attr since it doesn't allow types to have both "packed" and "aligned" attrs.
235  * So for Rust compatibility, don't include this. It can be included in the LE
236  * version because the "packed" attr is redundant in that case.
237  *
238  * History: (quoting Kent)
239  *
240  * Specifically, when i was designing bkey, I wanted the header to be no
241  * bigger than necessary so that bkey_packed could use the rest. That means that
242  * decently offten extent keys will fit into only 8 bytes, instead of spilling over
243  * to 16.
244  *
245  * But packed_bkey treats the part after the header - the packed section -
246  * as a single multi word, variable length integer. And bkey, the unpacked
247  * version, is just a special case version of a bkey_packed; all the packed
248  * bkey code will work on keys in any packed format, the in-memory
249  * representation of an unpacked key also is just one type of packed key...
250  *
251  * So that constrains the key part of a bkig endian bkey to start right
252  * after the header.
253  *
254  * If we ever do a bkey_v2 and need to expand the hedaer by another byte for
255  * some reason - that will clean up this wart.
256  */
257 __aligned(8)
258 #endif
259 ;
260 
261 struct bkey_packed {
262 	__u64		_data[0];
263 
264 	/* Size of combined key and value, in u64s */
265 	__u8		u64s;
266 
267 	/* Format of key (0 for format local to btree node) */
268 
269 	/*
270 	 * XXX: next incompat on disk format change, switch format and
271 	 * needs_whiteout - bkey_packed() will be cheaper if format is the high
272 	 * bits of the bitfield
273 	 */
274 #if defined(__LITTLE_ENDIAN_BITFIELD)
275 	__u8		format:7,
276 			needs_whiteout:1;
277 #elif defined (__BIG_ENDIAN_BITFIELD)
278 	__u8		needs_whiteout:1,
279 			format:7;
280 #endif
281 
282 	/* Type of the value */
283 	__u8		type;
284 	__u8		key_start[0];
285 
286 	/*
287 	 * We copy bkeys with struct assignment in various places, and while
288 	 * that shouldn't be done with packed bkeys we can't disallow it in C,
289 	 * and it's legal to cast a bkey to a bkey_packed  - so padding it out
290 	 * to the same size as struct bkey should hopefully be safest.
291 	 */
292 	__u8		pad[sizeof(struct bkey) - 3];
293 } __packed __aligned(8);
294 
295 typedef struct {
296 	__le64			lo;
297 	__le64			hi;
298 } bch_le128;
299 
300 #define BKEY_U64s			(sizeof(struct bkey) / sizeof(__u64))
301 #define BKEY_U64s_MAX			U8_MAX
302 #define BKEY_VAL_U64s_MAX		(BKEY_U64s_MAX - BKEY_U64s)
303 
304 #define KEY_PACKED_BITS_START		24
305 
306 #define KEY_FORMAT_LOCAL_BTREE		0
307 #define KEY_FORMAT_CURRENT		1
308 
309 enum bch_bkey_fields {
310 	BKEY_FIELD_INODE,
311 	BKEY_FIELD_OFFSET,
312 	BKEY_FIELD_SNAPSHOT,
313 	BKEY_FIELD_SIZE,
314 	BKEY_FIELD_VERSION_HI,
315 	BKEY_FIELD_VERSION_LO,
316 	BKEY_NR_FIELDS,
317 };
318 
319 #define bkey_format_field(name, field)					\
320 	[BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8)
321 
322 #define BKEY_FORMAT_CURRENT						\
323 ((struct bkey_format) {							\
324 	.key_u64s	= BKEY_U64s,					\
325 	.nr_fields	= BKEY_NR_FIELDS,				\
326 	.bits_per_field = {						\
327 		bkey_format_field(INODE,	p.inode),		\
328 		bkey_format_field(OFFSET,	p.offset),		\
329 		bkey_format_field(SNAPSHOT,	p.snapshot),		\
330 		bkey_format_field(SIZE,		size),			\
331 		bkey_format_field(VERSION_HI,	version.hi),		\
332 		bkey_format_field(VERSION_LO,	version.lo),		\
333 	},								\
334 })
335 
336 /* bkey with inline value */
337 struct bkey_i {
338 	__u64			_data[0];
339 
340 	struct bkey	k;
341 	struct bch_val	v;
342 };
343 
344 #define POS_KEY(_pos)							\
345 ((struct bkey) {							\
346 	.u64s		= BKEY_U64s,					\
347 	.format		= KEY_FORMAT_CURRENT,				\
348 	.p		= _pos,						\
349 })
350 
351 #define KEY(_inode, _offset, _size)					\
352 ((struct bkey) {							\
353 	.u64s		= BKEY_U64s,					\
354 	.format		= KEY_FORMAT_CURRENT,				\
355 	.p		= POS(_inode, _offset),				\
356 	.size		= _size,					\
357 })
358 
359 static inline void bkey_init(struct bkey *k)
360 {
361 	*k = KEY(0, 0, 0);
362 }
363 
364 #define bkey_bytes(_k)		((_k)->u64s * sizeof(__u64))
365 
366 #define __BKEY_PADDED(key, pad)					\
367 	struct bkey_i key; __u64 key ## _pad[pad]
368 
369 /*
370  * - DELETED keys are used internally to mark keys that should be ignored but
371  *   override keys in composition order.  Their version number is ignored.
372  *
373  * - DISCARDED keys indicate that the data is all 0s because it has been
374  *   discarded. DISCARDs may have a version; if the version is nonzero the key
375  *   will be persistent, otherwise the key will be dropped whenever the btree
376  *   node is rewritten (like DELETED keys).
377  *
378  * - ERROR: any read of the data returns a read error, as the data was lost due
379  *   to a failing device. Like DISCARDED keys, they can be removed (overridden)
380  *   by new writes or cluster-wide GC. Node repair can also overwrite them with
381  *   the same or a more recent version number, but not with an older version
382  *   number.
383  *
384  * - WHITEOUT: for hash table btrees
385  */
386 #define BCH_BKEY_TYPES()				\
387 	x(deleted,		0)			\
388 	x(whiteout,		1)			\
389 	x(error,		2)			\
390 	x(cookie,		3)			\
391 	x(hash_whiteout,	4)			\
392 	x(btree_ptr,		5)			\
393 	x(extent,		6)			\
394 	x(reservation,		7)			\
395 	x(inode,		8)			\
396 	x(inode_generation,	9)			\
397 	x(dirent,		10)			\
398 	x(xattr,		11)			\
399 	x(alloc,		12)			\
400 	x(quota,		13)			\
401 	x(stripe,		14)			\
402 	x(reflink_p,		15)			\
403 	x(reflink_v,		16)			\
404 	x(inline_data,		17)			\
405 	x(btree_ptr_v2,		18)			\
406 	x(indirect_inline_data,	19)			\
407 	x(alloc_v2,		20)			\
408 	x(subvolume,		21)			\
409 	x(snapshot,		22)			\
410 	x(inode_v2,		23)			\
411 	x(alloc_v3,		24)			\
412 	x(set,			25)			\
413 	x(lru,			26)			\
414 	x(alloc_v4,		27)			\
415 	x(backpointer,		28)			\
416 	x(inode_v3,		29)			\
417 	x(bucket_gens,		30)			\
418 	x(snapshot_tree,	31)			\
419 	x(logged_op_truncate,	32)			\
420 	x(logged_op_finsert,	33)
421 
422 enum bch_bkey_type {
423 #define x(name, nr) KEY_TYPE_##name	= nr,
424 	BCH_BKEY_TYPES()
425 #undef x
426 	KEY_TYPE_MAX,
427 };
428 
429 struct bch_deleted {
430 	struct bch_val		v;
431 };
432 
433 struct bch_whiteout {
434 	struct bch_val		v;
435 };
436 
437 struct bch_error {
438 	struct bch_val		v;
439 };
440 
441 struct bch_cookie {
442 	struct bch_val		v;
443 	__le64			cookie;
444 };
445 
446 struct bch_hash_whiteout {
447 	struct bch_val		v;
448 };
449 
450 struct bch_set {
451 	struct bch_val		v;
452 };
453 
454 /* 128 bits, sufficient for cryptographic MACs: */
455 struct bch_csum {
456 	__le64			lo;
457 	__le64			hi;
458 } __packed __aligned(8);
459 
460 struct bch_backpointer {
461 	struct bch_val		v;
462 	__u8			btree_id;
463 	__u8			level;
464 	__u8			data_type;
465 	__u64			bucket_offset:40;
466 	__u32			bucket_len;
467 	struct bpos		pos;
468 } __packed __aligned(8);
469 
470 /* LRU btree: */
471 
472 struct bch_lru {
473 	struct bch_val		v;
474 	__le64			idx;
475 } __packed __aligned(8);
476 
477 #define LRU_ID_STRIPES		(1U << 16)
478 
479 /* Optional/variable size superblock sections: */
480 
481 struct bch_sb_field {
482 	__u64			_data[0];
483 	__le32			u64s;
484 	__le32			type;
485 };
486 
487 #define BCH_SB_FIELDS()				\
488 	x(journal,			0)	\
489 	x(members_v1,			1)	\
490 	x(crypt,			2)	\
491 	x(replicas_v0,			3)	\
492 	x(quota,			4)	\
493 	x(disk_groups,			5)	\
494 	x(clean,			6)	\
495 	x(replicas,			7)	\
496 	x(journal_seq_blacklist,	8)	\
497 	x(journal_v2,			9)	\
498 	x(counters,			10)	\
499 	x(members_v2,			11)	\
500 	x(errors,			12)	\
501 	x(ext,				13)	\
502 	x(downgrade,			14)
503 
504 #include "alloc_background_format.h"
505 #include "extents_format.h"
506 #include "reflink_format.h"
507 #include "ec_format.h"
508 #include "inode_format.h"
509 #include "dirent_format.h"
510 #include "xattr_format.h"
511 #include "quota_format.h"
512 #include "logged_ops_format.h"
513 #include "snapshot_format.h"
514 #include "subvolume_format.h"
515 #include "sb-counters_format.h"
516 
517 enum bch_sb_field_type {
518 #define x(f, nr)	BCH_SB_FIELD_##f = nr,
519 	BCH_SB_FIELDS()
520 #undef x
521 	BCH_SB_FIELD_NR
522 };
523 
524 /*
525  * Most superblock fields are replicated in all device's superblocks - a few are
526  * not:
527  */
528 #define BCH_SINGLE_DEVICE_SB_FIELDS		\
529 	((1U << BCH_SB_FIELD_journal)|		\
530 	 (1U << BCH_SB_FIELD_journal_v2))
531 
532 /* BCH_SB_FIELD_journal: */
533 
534 struct bch_sb_field_journal {
535 	struct bch_sb_field	field;
536 	__le64			buckets[];
537 };
538 
539 struct bch_sb_field_journal_v2 {
540 	struct bch_sb_field	field;
541 
542 	struct bch_sb_field_journal_v2_entry {
543 		__le64		start;
544 		__le64		nr;
545 	}			d[];
546 };
547 
548 /* BCH_SB_FIELD_members_v1: */
549 
550 #define BCH_MIN_NR_NBUCKETS	(1 << 6)
551 
552 #define BCH_IOPS_MEASUREMENTS()			\
553 	x(seqread,	0)			\
554 	x(seqwrite,	1)			\
555 	x(randread,	2)			\
556 	x(randwrite,	3)
557 
558 enum bch_iops_measurement {
559 #define x(t, n) BCH_IOPS_##t = n,
560 	BCH_IOPS_MEASUREMENTS()
561 #undef x
562 	BCH_IOPS_NR
563 };
564 
565 #define BCH_MEMBER_ERROR_TYPES()		\
566 	x(read,		0)			\
567 	x(write,	1)			\
568 	x(checksum,	2)
569 
570 enum bch_member_error_type {
571 #define x(t, n) BCH_MEMBER_ERROR_##t = n,
572 	BCH_MEMBER_ERROR_TYPES()
573 #undef x
574 	BCH_MEMBER_ERROR_NR
575 };
576 
577 struct bch_member {
578 	__uuid_t		uuid;
579 	__le64			nbuckets;	/* device size */
580 	__le16			first_bucket;   /* index of first bucket used */
581 	__le16			bucket_size;	/* sectors */
582 	__u8			btree_bitmap_shift;
583 	__u8			pad[3];
584 	__le64			last_mount;	/* time_t */
585 
586 	__le64			flags;
587 	__le32			iops[4];
588 	__le64			errors[BCH_MEMBER_ERROR_NR];
589 	__le64			errors_at_reset[BCH_MEMBER_ERROR_NR];
590 	__le64			errors_reset_time;
591 	__le64			seq;
592 	__le64			btree_allocated_bitmap;
593 	/*
594 	 * On recovery from a clean shutdown we don't normally read the journal,
595 	 * but we still want to resume writing from where we left off so we
596 	 * don't overwrite more than is necessary, for list journal debugging:
597 	 */
598 	__le32			last_journal_bucket;
599 	__le32			last_journal_bucket_offset;
600 };
601 
602 /*
603  * This limit comes from the bucket_gens array - it's a single allocation, and
604  * kernel allocation are limited to INT_MAX
605  */
606 #define BCH_MEMBER_NBUCKETS_MAX	(INT_MAX - 64)
607 
608 #define BCH_MEMBER_V1_BYTES	56
609 
610 LE64_BITMASK(BCH_MEMBER_STATE,		struct bch_member, flags,  0,  4)
611 /* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */
612 LE64_BITMASK(BCH_MEMBER_DISCARD,	struct bch_member, flags, 14, 15)
613 LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED,	struct bch_member, flags, 15, 20)
614 LE64_BITMASK(BCH_MEMBER_GROUP,		struct bch_member, flags, 20, 28)
615 LE64_BITMASK(BCH_MEMBER_DURABILITY,	struct bch_member, flags, 28, 30)
616 LE64_BITMASK(BCH_MEMBER_FREESPACE_INITIALIZED,
617 					struct bch_member, flags, 30, 31)
618 
619 #if 0
620 LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS,	struct bch_member, flags[1], 0,  20);
621 LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40);
622 #endif
623 
624 #define BCH_MEMBER_STATES()			\
625 	x(rw,		0)			\
626 	x(ro,		1)			\
627 	x(failed,	2)			\
628 	x(spare,	3)
629 
630 enum bch_member_state {
631 #define x(t, n) BCH_MEMBER_STATE_##t = n,
632 	BCH_MEMBER_STATES()
633 #undef x
634 	BCH_MEMBER_STATE_NR
635 };
636 
637 struct bch_sb_field_members_v1 {
638 	struct bch_sb_field	field;
639 	struct bch_member	_members[]; //Members are now variable size
640 };
641 
642 struct bch_sb_field_members_v2 {
643 	struct bch_sb_field	field;
644 	__le16			member_bytes; //size of single member entry
645 	u8			pad[6];
646 	struct bch_member	_members[];
647 };
648 
649 /* BCH_SB_FIELD_crypt: */
650 
651 struct nonce {
652 	__le32			d[4];
653 };
654 
655 struct bch_key {
656 	__le64			key[4];
657 };
658 
659 #define BCH_KEY_MAGIC					\
660 	(((__u64) 'b' <<  0)|((__u64) 'c' <<  8)|		\
661 	 ((__u64) 'h' << 16)|((__u64) '*' << 24)|		\
662 	 ((__u64) '*' << 32)|((__u64) 'k' << 40)|		\
663 	 ((__u64) 'e' << 48)|((__u64) 'y' << 56))
664 
665 struct bch_encrypted_key {
666 	__le64			magic;
667 	struct bch_key		key;
668 };
669 
670 /*
671  * If this field is present in the superblock, it stores an encryption key which
672  * is used encrypt all other data/metadata. The key will normally be encrypted
673  * with the key userspace provides, but if encryption has been turned off we'll
674  * just store the master key unencrypted in the superblock so we can access the
675  * previously encrypted data.
676  */
677 struct bch_sb_field_crypt {
678 	struct bch_sb_field	field;
679 
680 	__le64			flags;
681 	__le64			kdf_flags;
682 	struct bch_encrypted_key key;
683 };
684 
685 LE64_BITMASK(BCH_CRYPT_KDF_TYPE,	struct bch_sb_field_crypt, flags, 0, 4);
686 
687 enum bch_kdf_types {
688 	BCH_KDF_SCRYPT		= 0,
689 	BCH_KDF_NR		= 1,
690 };
691 
692 /* stored as base 2 log of scrypt params: */
693 LE64_BITMASK(BCH_KDF_SCRYPT_N,	struct bch_sb_field_crypt, kdf_flags,  0, 16);
694 LE64_BITMASK(BCH_KDF_SCRYPT_R,	struct bch_sb_field_crypt, kdf_flags, 16, 32);
695 LE64_BITMASK(BCH_KDF_SCRYPT_P,	struct bch_sb_field_crypt, kdf_flags, 32, 48);
696 
697 /* BCH_SB_FIELD_replicas: */
698 
699 #define BCH_DATA_TYPES()		\
700 	x(free,		0)		\
701 	x(sb,		1)		\
702 	x(journal,	2)		\
703 	x(btree,	3)		\
704 	x(user,		4)		\
705 	x(cached,	5)		\
706 	x(parity,	6)		\
707 	x(stripe,	7)		\
708 	x(need_gc_gens,	8)		\
709 	x(need_discard,	9)
710 
711 enum bch_data_type {
712 #define x(t, n) BCH_DATA_##t,
713 	BCH_DATA_TYPES()
714 #undef x
715 	BCH_DATA_NR
716 };
717 
718 static inline bool data_type_is_empty(enum bch_data_type type)
719 {
720 	switch (type) {
721 	case BCH_DATA_free:
722 	case BCH_DATA_need_gc_gens:
723 	case BCH_DATA_need_discard:
724 		return true;
725 	default:
726 		return false;
727 	}
728 }
729 
730 static inline bool data_type_is_hidden(enum bch_data_type type)
731 {
732 	switch (type) {
733 	case BCH_DATA_sb:
734 	case BCH_DATA_journal:
735 		return true;
736 	default:
737 		return false;
738 	}
739 }
740 
741 struct bch_replicas_entry_v0 {
742 	__u8			data_type;
743 	__u8			nr_devs;
744 	__u8			devs[];
745 } __packed;
746 
747 struct bch_sb_field_replicas_v0 {
748 	struct bch_sb_field	field;
749 	struct bch_replicas_entry_v0 entries[];
750 } __packed __aligned(8);
751 
752 struct bch_replicas_entry_v1 {
753 	__u8			data_type;
754 	__u8			nr_devs;
755 	__u8			nr_required;
756 	__u8			devs[];
757 } __packed;
758 
759 #define replicas_entry_bytes(_i)					\
760 	(offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
761 
762 struct bch_sb_field_replicas {
763 	struct bch_sb_field	field;
764 	struct bch_replicas_entry_v1 entries[];
765 } __packed __aligned(8);
766 
767 /* BCH_SB_FIELD_disk_groups: */
768 
769 #define BCH_SB_LABEL_SIZE		32
770 
771 struct bch_disk_group {
772 	__u8			label[BCH_SB_LABEL_SIZE];
773 	__le64			flags[2];
774 } __packed __aligned(8);
775 
776 LE64_BITMASK(BCH_GROUP_DELETED,		struct bch_disk_group, flags[0], 0,  1)
777 LE64_BITMASK(BCH_GROUP_DATA_ALLOWED,	struct bch_disk_group, flags[0], 1,  6)
778 LE64_BITMASK(BCH_GROUP_PARENT,		struct bch_disk_group, flags[0], 6, 24)
779 
780 struct bch_sb_field_disk_groups {
781 	struct bch_sb_field	field;
782 	struct bch_disk_group	entries[];
783 } __packed __aligned(8);
784 
785 /*
786  * On clean shutdown, store btree roots and current journal sequence number in
787  * the superblock:
788  */
789 struct jset_entry {
790 	__le16			u64s;
791 	__u8			btree_id;
792 	__u8			level;
793 	__u8			type; /* designates what this jset holds */
794 	__u8			pad[3];
795 
796 	struct bkey_i		start[0];
797 	__u64			_data[];
798 };
799 
800 struct bch_sb_field_clean {
801 	struct bch_sb_field	field;
802 
803 	__le32			flags;
804 	__le16			_read_clock; /* no longer used */
805 	__le16			_write_clock;
806 	__le64			journal_seq;
807 
808 	struct jset_entry	start[0];
809 	__u64			_data[];
810 };
811 
812 struct journal_seq_blacklist_entry {
813 	__le64			start;
814 	__le64			end;
815 };
816 
817 struct bch_sb_field_journal_seq_blacklist {
818 	struct bch_sb_field	field;
819 	struct journal_seq_blacklist_entry start[];
820 };
821 
822 struct bch_sb_field_errors {
823 	struct bch_sb_field	field;
824 	struct bch_sb_field_error_entry {
825 		__le64		v;
826 		__le64		last_error_time;
827 	}			entries[];
828 };
829 
830 LE64_BITMASK(BCH_SB_ERROR_ENTRY_ID,	struct bch_sb_field_error_entry, v,  0, 16);
831 LE64_BITMASK(BCH_SB_ERROR_ENTRY_NR,	struct bch_sb_field_error_entry, v, 16, 64);
832 
833 struct bch_sb_field_ext {
834 	struct bch_sb_field	field;
835 	__le64			recovery_passes_required[2];
836 	__le64			errors_silent[8];
837 	__le64			btrees_lost_data;
838 };
839 
840 struct bch_sb_field_downgrade_entry {
841 	__le16			version;
842 	__le64			recovery_passes[2];
843 	__le16			nr_errors;
844 	__le16			errors[] __counted_by(nr_errors);
845 } __packed __aligned(2);
846 
847 struct bch_sb_field_downgrade {
848 	struct bch_sb_field	field;
849 	struct bch_sb_field_downgrade_entry entries[];
850 };
851 
852 /* Superblock: */
853 
854 /*
855  * New versioning scheme:
856  * One common version number for all on disk data structures - superblock, btree
857  * nodes, journal entries
858  */
859 #define BCH_VERSION_MAJOR(_v)		((__u16) ((_v) >> 10))
860 #define BCH_VERSION_MINOR(_v)		((__u16) ((_v) & ~(~0U << 10)))
861 #define BCH_VERSION(_major, _minor)	(((_major) << 10)|(_minor) << 0)
862 
863 /*
864  * field 1:		version name
865  * field 2:		BCH_VERSION(major, minor)
866  * field 3:		recovery passess required on upgrade
867  */
868 #define BCH_METADATA_VERSIONS()						\
869 	x(bkey_renumber,		BCH_VERSION(0, 10))		\
870 	x(inode_btree_change,		BCH_VERSION(0, 11))		\
871 	x(snapshot,			BCH_VERSION(0, 12))		\
872 	x(inode_backpointers,		BCH_VERSION(0, 13))		\
873 	x(btree_ptr_sectors_written,	BCH_VERSION(0, 14))		\
874 	x(snapshot_2,			BCH_VERSION(0, 15))		\
875 	x(reflink_p_fix,		BCH_VERSION(0, 16))		\
876 	x(subvol_dirent,		BCH_VERSION(0, 17))		\
877 	x(inode_v2,			BCH_VERSION(0, 18))		\
878 	x(freespace,			BCH_VERSION(0, 19))		\
879 	x(alloc_v4,			BCH_VERSION(0, 20))		\
880 	x(new_data_types,		BCH_VERSION(0, 21))		\
881 	x(backpointers,			BCH_VERSION(0, 22))		\
882 	x(inode_v3,			BCH_VERSION(0, 23))		\
883 	x(unwritten_extents,		BCH_VERSION(0, 24))		\
884 	x(bucket_gens,			BCH_VERSION(0, 25))		\
885 	x(lru_v2,			BCH_VERSION(0, 26))		\
886 	x(fragmentation_lru,		BCH_VERSION(0, 27))		\
887 	x(no_bps_in_alloc_keys,		BCH_VERSION(0, 28))		\
888 	x(snapshot_trees,		BCH_VERSION(0, 29))		\
889 	x(major_minor,			BCH_VERSION(1,  0))		\
890 	x(snapshot_skiplists,		BCH_VERSION(1,  1))		\
891 	x(deleted_inodes,		BCH_VERSION(1,  2))		\
892 	x(rebalance_work,		BCH_VERSION(1,  3))		\
893 	x(member_seq,			BCH_VERSION(1,  4))		\
894 	x(subvolume_fs_parent,		BCH_VERSION(1,  5))		\
895 	x(btree_subvolume_children,	BCH_VERSION(1,  6))		\
896 	x(mi_btree_bitmap,		BCH_VERSION(1,  7))
897 
898 enum bcachefs_metadata_version {
899 	bcachefs_metadata_version_min = 9,
900 #define x(t, n)	bcachefs_metadata_version_##t = n,
901 	BCH_METADATA_VERSIONS()
902 #undef x
903 	bcachefs_metadata_version_max
904 };
905 
906 static const __maybe_unused
907 unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
908 
909 #define bcachefs_metadata_version_current	(bcachefs_metadata_version_max - 1)
910 
911 #define BCH_SB_SECTOR			8
912 #define BCH_SB_MEMBERS_MAX		64 /* XXX kill */
913 
914 #define BCH_SB_LAYOUT_SIZE_BITS_MAX	16 /* 32 MB */
915 
916 struct bch_sb_layout {
917 	__uuid_t		magic;	/* bcachefs superblock UUID */
918 	__u8			layout_type;
919 	__u8			sb_max_size_bits; /* base 2 of 512 byte sectors */
920 	__u8			nr_superblocks;
921 	__u8			pad[5];
922 	__le64			sb_offset[61];
923 } __packed __aligned(8);
924 
925 #define BCH_SB_LAYOUT_SECTOR	7
926 
927 /*
928  * @offset	- sector where this sb was written
929  * @version	- on disk format version
930  * @version_min	- Oldest metadata version this filesystem contains; so we can
931  *		  safely drop compatibility code and refuse to mount filesystems
932  *		  we'd need it for
933  * @magic	- identifies as a bcachefs superblock (BCHFS_MAGIC)
934  * @seq		- incremented each time superblock is written
935  * @uuid	- used for generating various magic numbers and identifying
936  *                member devices, never changes
937  * @user_uuid	- user visible UUID, may be changed
938  * @label	- filesystem label
939  * @seq		- identifies most recent superblock, incremented each time
940  *		  superblock is written
941  * @features	- enabled incompatible features
942  */
943 struct bch_sb {
944 	struct bch_csum		csum;
945 	__le16			version;
946 	__le16			version_min;
947 	__le16			pad[2];
948 	__uuid_t		magic;
949 	__uuid_t		uuid;
950 	__uuid_t		user_uuid;
951 	__u8			label[BCH_SB_LABEL_SIZE];
952 	__le64			offset;
953 	__le64			seq;
954 
955 	__le16			block_size;
956 	__u8			dev_idx;
957 	__u8			nr_devices;
958 	__le32			u64s;
959 
960 	__le64			time_base_lo;
961 	__le32			time_base_hi;
962 	__le32			time_precision;
963 
964 	__le64			flags[7];
965 	__le64			write_time;
966 	__le64			features[2];
967 	__le64			compat[2];
968 
969 	struct bch_sb_layout	layout;
970 
971 	struct bch_sb_field	start[0];
972 	__le64			_data[];
973 } __packed __aligned(8);
974 
975 /*
976  * Flags:
977  * BCH_SB_INITALIZED	- set on first mount
978  * BCH_SB_CLEAN		- did we shut down cleanly? Just a hint, doesn't affect
979  *			  behaviour of mount/recovery path:
980  * BCH_SB_INODE_32BIT	- limit inode numbers to 32 bits
981  * BCH_SB_128_BIT_MACS	- 128 bit macs instead of 80
982  * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides
983  *			   DATA/META_CSUM_TYPE. Also indicates encryption
984  *			   algorithm in use, if/when we get more than one
985  */
986 
987 LE16_BITMASK(BCH_SB_BLOCK_SIZE,		struct bch_sb, block_size, 0, 16);
988 
989 LE64_BITMASK(BCH_SB_INITIALIZED,	struct bch_sb, flags[0],  0,  1);
990 LE64_BITMASK(BCH_SB_CLEAN,		struct bch_sb, flags[0],  1,  2);
991 LE64_BITMASK(BCH_SB_CSUM_TYPE,		struct bch_sb, flags[0],  2,  8);
992 LE64_BITMASK(BCH_SB_ERROR_ACTION,	struct bch_sb, flags[0],  8, 12);
993 
994 LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE,	struct bch_sb, flags[0], 12, 28);
995 
996 LE64_BITMASK(BCH_SB_GC_RESERVE,		struct bch_sb, flags[0], 28, 33);
997 LE64_BITMASK(BCH_SB_ROOT_RESERVE,	struct bch_sb, flags[0], 33, 40);
998 
999 LE64_BITMASK(BCH_SB_META_CSUM_TYPE,	struct bch_sb, flags[0], 40, 44);
1000 LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE,	struct bch_sb, flags[0], 44, 48);
1001 
1002 LE64_BITMASK(BCH_SB_META_REPLICAS_WANT,	struct bch_sb, flags[0], 48, 52);
1003 LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT,	struct bch_sb, flags[0], 52, 56);
1004 
1005 LE64_BITMASK(BCH_SB_POSIX_ACL,		struct bch_sb, flags[0], 56, 57);
1006 LE64_BITMASK(BCH_SB_USRQUOTA,		struct bch_sb, flags[0], 57, 58);
1007 LE64_BITMASK(BCH_SB_GRPQUOTA,		struct bch_sb, flags[0], 58, 59);
1008 LE64_BITMASK(BCH_SB_PRJQUOTA,		struct bch_sb, flags[0], 59, 60);
1009 
1010 LE64_BITMASK(BCH_SB_HAS_ERRORS,		struct bch_sb, flags[0], 60, 61);
1011 LE64_BITMASK(BCH_SB_HAS_TOPOLOGY_ERRORS,struct bch_sb, flags[0], 61, 62);
1012 
1013 LE64_BITMASK(BCH_SB_BIG_ENDIAN,		struct bch_sb, flags[0], 62, 63);
1014 
1015 LE64_BITMASK(BCH_SB_STR_HASH_TYPE,	struct bch_sb, flags[1],  0,  4);
1016 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_LO,struct bch_sb, flags[1],  4,  8);
1017 LE64_BITMASK(BCH_SB_INODE_32BIT,	struct bch_sb, flags[1],  8,  9);
1018 
1019 LE64_BITMASK(BCH_SB_128_BIT_MACS,	struct bch_sb, flags[1],  9, 10);
1020 LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE,	struct bch_sb, flags[1], 10, 14);
1021 
1022 /*
1023  * Max size of an extent that may require bouncing to read or write
1024  * (checksummed, compressed): 64k
1025  */
1026 LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS,
1027 					struct bch_sb, flags[1], 14, 20);
1028 
1029 LE64_BITMASK(BCH_SB_META_REPLICAS_REQ,	struct bch_sb, flags[1], 20, 24);
1030 LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ,	struct bch_sb, flags[1], 24, 28);
1031 
1032 LE64_BITMASK(BCH_SB_PROMOTE_TARGET,	struct bch_sb, flags[1], 28, 40);
1033 LE64_BITMASK(BCH_SB_FOREGROUND_TARGET,	struct bch_sb, flags[1], 40, 52);
1034 LE64_BITMASK(BCH_SB_BACKGROUND_TARGET,	struct bch_sb, flags[1], 52, 64);
1035 
1036 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO,
1037 					struct bch_sb, flags[2],  0,  4);
1038 LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES,	struct bch_sb, flags[2],  4, 64);
1039 
1040 LE64_BITMASK(BCH_SB_ERASURE_CODE,	struct bch_sb, flags[3],  0, 16);
1041 LE64_BITMASK(BCH_SB_METADATA_TARGET,	struct bch_sb, flags[3], 16, 28);
1042 LE64_BITMASK(BCH_SB_SHARD_INUMS,	struct bch_sb, flags[3], 28, 29);
1043 LE64_BITMASK(BCH_SB_INODES_USE_KEY_CACHE,struct bch_sb, flags[3], 29, 30);
1044 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DELAY,struct bch_sb, flags[3], 30, 62);
1045 LE64_BITMASK(BCH_SB_JOURNAL_FLUSH_DISABLED,struct bch_sb, flags[3], 62, 63);
1046 LE64_BITMASK(BCH_SB_JOURNAL_RECLAIM_DELAY,struct bch_sb, flags[4], 0, 32);
1047 LE64_BITMASK(BCH_SB_JOURNAL_TRANSACTION_NAMES,struct bch_sb, flags[4], 32, 33);
1048 LE64_BITMASK(BCH_SB_NOCOW,		struct bch_sb, flags[4], 33, 34);
1049 LE64_BITMASK(BCH_SB_WRITE_BUFFER_SIZE,	struct bch_sb, flags[4], 34, 54);
1050 LE64_BITMASK(BCH_SB_VERSION_UPGRADE,	struct bch_sb, flags[4], 54, 56);
1051 
1052 LE64_BITMASK(BCH_SB_COMPRESSION_TYPE_HI,struct bch_sb, flags[4], 56, 60);
1053 LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI,
1054 					struct bch_sb, flags[4], 60, 64);
1055 
1056 LE64_BITMASK(BCH_SB_VERSION_UPGRADE_COMPLETE,
1057 					struct bch_sb, flags[5],  0, 16);
1058 
1059 static inline __u64 BCH_SB_COMPRESSION_TYPE(const struct bch_sb *sb)
1060 {
1061 	return BCH_SB_COMPRESSION_TYPE_LO(sb) | (BCH_SB_COMPRESSION_TYPE_HI(sb) << 4);
1062 }
1063 
1064 static inline void SET_BCH_SB_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1065 {
1066 	SET_BCH_SB_COMPRESSION_TYPE_LO(sb, v);
1067 	SET_BCH_SB_COMPRESSION_TYPE_HI(sb, v >> 4);
1068 }
1069 
1070 static inline __u64 BCH_SB_BACKGROUND_COMPRESSION_TYPE(const struct bch_sb *sb)
1071 {
1072 	return BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb) |
1073 		(BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb) << 4);
1074 }
1075 
1076 static inline void SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE(struct bch_sb *sb, __u64 v)
1077 {
1078 	SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_LO(sb, v);
1079 	SET_BCH_SB_BACKGROUND_COMPRESSION_TYPE_HI(sb, v >> 4);
1080 }
1081 
1082 /*
1083  * Features:
1084  *
1085  * journal_seq_blacklist_v3:	gates BCH_SB_FIELD_journal_seq_blacklist
1086  * reflink:			gates KEY_TYPE_reflink
1087  * inline_data:			gates KEY_TYPE_inline_data
1088  * new_siphash:			gates BCH_STR_HASH_siphash
1089  * new_extent_overwrite:	gates BTREE_NODE_NEW_EXTENT_OVERWRITE
1090  */
1091 #define BCH_SB_FEATURES()			\
1092 	x(lz4,				0)	\
1093 	x(gzip,				1)	\
1094 	x(zstd,				2)	\
1095 	x(atomic_nlink,			3)	\
1096 	x(ec,				4)	\
1097 	x(journal_seq_blacklist_v3,	5)	\
1098 	x(reflink,			6)	\
1099 	x(new_siphash,			7)	\
1100 	x(inline_data,			8)	\
1101 	x(new_extent_overwrite,		9)	\
1102 	x(incompressible,		10)	\
1103 	x(btree_ptr_v2,			11)	\
1104 	x(extents_above_btree_updates,	12)	\
1105 	x(btree_updates_journalled,	13)	\
1106 	x(reflink_inline_data,		14)	\
1107 	x(new_varint,			15)	\
1108 	x(journal_no_flush,		16)	\
1109 	x(alloc_v2,			17)	\
1110 	x(extents_across_btree_nodes,	18)
1111 
1112 #define BCH_SB_FEATURES_ALWAYS				\
1113 	((1ULL << BCH_FEATURE_new_extent_overwrite)|	\
1114 	 (1ULL << BCH_FEATURE_extents_above_btree_updates)|\
1115 	 (1ULL << BCH_FEATURE_btree_updates_journalled)|\
1116 	 (1ULL << BCH_FEATURE_alloc_v2)|\
1117 	 (1ULL << BCH_FEATURE_extents_across_btree_nodes))
1118 
1119 #define BCH_SB_FEATURES_ALL				\
1120 	(BCH_SB_FEATURES_ALWAYS|			\
1121 	 (1ULL << BCH_FEATURE_new_siphash)|		\
1122 	 (1ULL << BCH_FEATURE_btree_ptr_v2)|		\
1123 	 (1ULL << BCH_FEATURE_new_varint)|		\
1124 	 (1ULL << BCH_FEATURE_journal_no_flush))
1125 
1126 enum bch_sb_feature {
1127 #define x(f, n) BCH_FEATURE_##f,
1128 	BCH_SB_FEATURES()
1129 #undef x
1130 	BCH_FEATURE_NR,
1131 };
1132 
1133 #define BCH_SB_COMPAT()					\
1134 	x(alloc_info,				0)	\
1135 	x(alloc_metadata,			1)	\
1136 	x(extents_above_btree_updates_done,	2)	\
1137 	x(bformat_overflow_done,		3)
1138 
1139 enum bch_sb_compat {
1140 #define x(f, n) BCH_COMPAT_##f,
1141 	BCH_SB_COMPAT()
1142 #undef x
1143 	BCH_COMPAT_NR,
1144 };
1145 
1146 /* options: */
1147 
1148 #define BCH_VERSION_UPGRADE_OPTS()	\
1149 	x(compatible,		0)	\
1150 	x(incompatible,		1)	\
1151 	x(none,			2)
1152 
1153 enum bch_version_upgrade_opts {
1154 #define x(t, n) BCH_VERSION_UPGRADE_##t = n,
1155 	BCH_VERSION_UPGRADE_OPTS()
1156 #undef x
1157 };
1158 
1159 #define BCH_REPLICAS_MAX		4U
1160 
1161 #define BCH_BKEY_PTRS_MAX		16U
1162 
1163 #define BCH_ERROR_ACTIONS()		\
1164 	x(continue,		0)	\
1165 	x(ro,			1)	\
1166 	x(panic,		2)
1167 
1168 enum bch_error_actions {
1169 #define x(t, n) BCH_ON_ERROR_##t = n,
1170 	BCH_ERROR_ACTIONS()
1171 #undef x
1172 	BCH_ON_ERROR_NR
1173 };
1174 
1175 #define BCH_STR_HASH_TYPES()		\
1176 	x(crc32c,		0)	\
1177 	x(crc64,		1)	\
1178 	x(siphash_old,		2)	\
1179 	x(siphash,		3)
1180 
1181 enum bch_str_hash_type {
1182 #define x(t, n) BCH_STR_HASH_##t = n,
1183 	BCH_STR_HASH_TYPES()
1184 #undef x
1185 	BCH_STR_HASH_NR
1186 };
1187 
1188 #define BCH_STR_HASH_OPTS()		\
1189 	x(crc32c,		0)	\
1190 	x(crc64,		1)	\
1191 	x(siphash,		2)
1192 
1193 enum bch_str_hash_opts {
1194 #define x(t, n) BCH_STR_HASH_OPT_##t = n,
1195 	BCH_STR_HASH_OPTS()
1196 #undef x
1197 	BCH_STR_HASH_OPT_NR
1198 };
1199 
1200 #define BCH_CSUM_TYPES()			\
1201 	x(none,				0)	\
1202 	x(crc32c_nonzero,		1)	\
1203 	x(crc64_nonzero,		2)	\
1204 	x(chacha20_poly1305_80,		3)	\
1205 	x(chacha20_poly1305_128,	4)	\
1206 	x(crc32c,			5)	\
1207 	x(crc64,			6)	\
1208 	x(xxhash,			7)
1209 
1210 enum bch_csum_type {
1211 #define x(t, n) BCH_CSUM_##t = n,
1212 	BCH_CSUM_TYPES()
1213 #undef x
1214 	BCH_CSUM_NR
1215 };
1216 
1217 static const __maybe_unused unsigned bch_crc_bytes[] = {
1218 	[BCH_CSUM_none]				= 0,
1219 	[BCH_CSUM_crc32c_nonzero]		= 4,
1220 	[BCH_CSUM_crc32c]			= 4,
1221 	[BCH_CSUM_crc64_nonzero]		= 8,
1222 	[BCH_CSUM_crc64]			= 8,
1223 	[BCH_CSUM_xxhash]			= 8,
1224 	[BCH_CSUM_chacha20_poly1305_80]		= 10,
1225 	[BCH_CSUM_chacha20_poly1305_128]	= 16,
1226 };
1227 
1228 static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
1229 {
1230 	switch (type) {
1231 	case BCH_CSUM_chacha20_poly1305_80:
1232 	case BCH_CSUM_chacha20_poly1305_128:
1233 		return true;
1234 	default:
1235 		return false;
1236 	}
1237 }
1238 
1239 #define BCH_CSUM_OPTS()			\
1240 	x(none,			0)	\
1241 	x(crc32c,		1)	\
1242 	x(crc64,		2)	\
1243 	x(xxhash,		3)
1244 
1245 enum bch_csum_opts {
1246 #define x(t, n) BCH_CSUM_OPT_##t = n,
1247 	BCH_CSUM_OPTS()
1248 #undef x
1249 	BCH_CSUM_OPT_NR
1250 };
1251 
1252 #define BCH_COMPRESSION_TYPES()		\
1253 	x(none,			0)	\
1254 	x(lz4_old,		1)	\
1255 	x(gzip,			2)	\
1256 	x(lz4,			3)	\
1257 	x(zstd,			4)	\
1258 	x(incompressible,	5)
1259 
1260 enum bch_compression_type {
1261 #define x(t, n) BCH_COMPRESSION_TYPE_##t = n,
1262 	BCH_COMPRESSION_TYPES()
1263 #undef x
1264 	BCH_COMPRESSION_TYPE_NR
1265 };
1266 
1267 #define BCH_COMPRESSION_OPTS()		\
1268 	x(none,		0)		\
1269 	x(lz4,		1)		\
1270 	x(gzip,		2)		\
1271 	x(zstd,		3)
1272 
1273 enum bch_compression_opts {
1274 #define x(t, n) BCH_COMPRESSION_OPT_##t = n,
1275 	BCH_COMPRESSION_OPTS()
1276 #undef x
1277 	BCH_COMPRESSION_OPT_NR
1278 };
1279 
1280 /*
1281  * Magic numbers
1282  *
1283  * The various other data structures have their own magic numbers, which are
1284  * xored with the first part of the cache set's UUID
1285  */
1286 
1287 #define BCACHE_MAGIC							\
1288 	UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca,				\
1289 		  0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81)
1290 #define BCHFS_MAGIC							\
1291 	UUID_INIT(0xc68573f6, 0x66ce, 0x90a9,				\
1292 		  0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef)
1293 
1294 #define BCACHEFS_STATFS_MAGIC		BCACHEFS_SUPER_MAGIC
1295 
1296 #define JSET_MAGIC		__cpu_to_le64(0x245235c1a3625032ULL)
1297 #define BSET_MAGIC		__cpu_to_le64(0x90135c78b99e07f5ULL)
1298 
1299 static inline __le64 __bch2_sb_magic(struct bch_sb *sb)
1300 {
1301 	__le64 ret;
1302 
1303 	memcpy(&ret, &sb->uuid, sizeof(ret));
1304 	return ret;
1305 }
1306 
1307 static inline __u64 __jset_magic(struct bch_sb *sb)
1308 {
1309 	return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC);
1310 }
1311 
1312 static inline __u64 __bset_magic(struct bch_sb *sb)
1313 {
1314 	return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC);
1315 }
1316 
1317 /* Journal */
1318 
1319 #define JSET_KEYS_U64s	(sizeof(struct jset_entry) / sizeof(__u64))
1320 
1321 #define BCH_JSET_ENTRY_TYPES()			\
1322 	x(btree_keys,		0)		\
1323 	x(btree_root,		1)		\
1324 	x(prio_ptrs,		2)		\
1325 	x(blacklist,		3)		\
1326 	x(blacklist_v2,		4)		\
1327 	x(usage,		5)		\
1328 	x(data_usage,		6)		\
1329 	x(clock,		7)		\
1330 	x(dev_usage,		8)		\
1331 	x(log,			9)		\
1332 	x(overwrite,		10)		\
1333 	x(write_buffer_keys,	11)		\
1334 	x(datetime,		12)
1335 
1336 enum bch_jset_entry_type {
1337 #define x(f, nr)	BCH_JSET_ENTRY_##f	= nr,
1338 	BCH_JSET_ENTRY_TYPES()
1339 #undef x
1340 	BCH_JSET_ENTRY_NR
1341 };
1342 
1343 static inline bool jset_entry_is_key(struct jset_entry *e)
1344 {
1345 	switch (e->type) {
1346 	case BCH_JSET_ENTRY_btree_keys:
1347 	case BCH_JSET_ENTRY_btree_root:
1348 	case BCH_JSET_ENTRY_overwrite:
1349 	case BCH_JSET_ENTRY_write_buffer_keys:
1350 		return true;
1351 	}
1352 
1353 	return false;
1354 }
1355 
1356 /*
1357  * Journal sequence numbers can be blacklisted: bsets record the max sequence
1358  * number of all the journal entries they contain updates for, so that on
1359  * recovery we can ignore those bsets that contain index updates newer that what
1360  * made it into the journal.
1361  *
1362  * This means that we can't reuse that journal_seq - we have to skip it, and
1363  * then record that we skipped it so that the next time we crash and recover we
1364  * don't think there was a missing journal entry.
1365  */
1366 struct jset_entry_blacklist {
1367 	struct jset_entry	entry;
1368 	__le64			seq;
1369 };
1370 
1371 struct jset_entry_blacklist_v2 {
1372 	struct jset_entry	entry;
1373 	__le64			start;
1374 	__le64			end;
1375 };
1376 
1377 #define BCH_FS_USAGE_TYPES()			\
1378 	x(reserved,		0)		\
1379 	x(inodes,		1)		\
1380 	x(key_version,		2)
1381 
1382 enum bch_fs_usage_type {
1383 #define x(f, nr)	BCH_FS_USAGE_##f	= nr,
1384 	BCH_FS_USAGE_TYPES()
1385 #undef x
1386 	BCH_FS_USAGE_NR
1387 };
1388 
1389 struct jset_entry_usage {
1390 	struct jset_entry	entry;
1391 	__le64			v;
1392 } __packed;
1393 
1394 struct jset_entry_data_usage {
1395 	struct jset_entry	entry;
1396 	__le64			v;
1397 	struct bch_replicas_entry_v1 r;
1398 } __packed;
1399 
1400 struct jset_entry_clock {
1401 	struct jset_entry	entry;
1402 	__u8			rw;
1403 	__u8			pad[7];
1404 	__le64			time;
1405 } __packed;
1406 
1407 struct jset_entry_dev_usage_type {
1408 	__le64			buckets;
1409 	__le64			sectors;
1410 	__le64			fragmented;
1411 } __packed;
1412 
1413 struct jset_entry_dev_usage {
1414 	struct jset_entry	entry;
1415 	__le32			dev;
1416 	__u32			pad;
1417 
1418 	__le64			_buckets_ec;		/* No longer used */
1419 	__le64			_buckets_unavailable;	/* No longer used */
1420 
1421 	struct jset_entry_dev_usage_type d[];
1422 };
1423 
1424 static inline unsigned jset_entry_dev_usage_nr_types(struct jset_entry_dev_usage *u)
1425 {
1426 	return (vstruct_bytes(&u->entry) - sizeof(struct jset_entry_dev_usage)) /
1427 		sizeof(struct jset_entry_dev_usage_type);
1428 }
1429 
1430 struct jset_entry_log {
1431 	struct jset_entry	entry;
1432 	u8			d[];
1433 } __packed __aligned(8);
1434 
1435 struct jset_entry_datetime {
1436 	struct jset_entry	entry;
1437 	__le64			seconds;
1438 } __packed __aligned(8);
1439 
1440 /*
1441  * On disk format for a journal entry:
1442  * seq is monotonically increasing; every journal entry has its own unique
1443  * sequence number.
1444  *
1445  * last_seq is the oldest journal entry that still has keys the btree hasn't
1446  * flushed to disk yet.
1447  *
1448  * version is for on disk format changes.
1449  */
1450 struct jset {
1451 	struct bch_csum		csum;
1452 
1453 	__le64			magic;
1454 	__le64			seq;
1455 	__le32			version;
1456 	__le32			flags;
1457 
1458 	__le32			u64s; /* size of d[] in u64s */
1459 
1460 	__u8			encrypted_start[0];
1461 
1462 	__le16			_read_clock; /* no longer used */
1463 	__le16			_write_clock;
1464 
1465 	/* Sequence number of oldest dirty journal entry */
1466 	__le64			last_seq;
1467 
1468 
1469 	struct jset_entry	start[0];
1470 	__u64			_data[];
1471 } __packed __aligned(8);
1472 
1473 LE32_BITMASK(JSET_CSUM_TYPE,	struct jset, flags, 0, 4);
1474 LE32_BITMASK(JSET_BIG_ENDIAN,	struct jset, flags, 4, 5);
1475 LE32_BITMASK(JSET_NO_FLUSH,	struct jset, flags, 5, 6);
1476 
1477 #define BCH_JOURNAL_BUCKETS_MIN		8
1478 
1479 /* Btree: */
1480 
1481 enum btree_id_flags {
1482 	BTREE_ID_EXTENTS	= BIT(0),
1483 	BTREE_ID_SNAPSHOTS	= BIT(1),
1484 	BTREE_ID_SNAPSHOT_FIELD	= BIT(2),
1485 	BTREE_ID_DATA		= BIT(3),
1486 };
1487 
1488 #define BCH_BTREE_IDS()								\
1489 	x(extents,		0,	BTREE_ID_EXTENTS|BTREE_ID_SNAPSHOTS|BTREE_ID_DATA,\
1490 	  BIT_ULL(KEY_TYPE_whiteout)|						\
1491 	  BIT_ULL(KEY_TYPE_error)|						\
1492 	  BIT_ULL(KEY_TYPE_cookie)|						\
1493 	  BIT_ULL(KEY_TYPE_extent)|						\
1494 	  BIT_ULL(KEY_TYPE_reservation)|					\
1495 	  BIT_ULL(KEY_TYPE_reflink_p)|						\
1496 	  BIT_ULL(KEY_TYPE_inline_data))					\
1497 	x(inodes,		1,	BTREE_ID_SNAPSHOTS,			\
1498 	  BIT_ULL(KEY_TYPE_whiteout)|						\
1499 	  BIT_ULL(KEY_TYPE_inode)|						\
1500 	  BIT_ULL(KEY_TYPE_inode_v2)|						\
1501 	  BIT_ULL(KEY_TYPE_inode_v3)|						\
1502 	  BIT_ULL(KEY_TYPE_inode_generation))					\
1503 	x(dirents,		2,	BTREE_ID_SNAPSHOTS,			\
1504 	  BIT_ULL(KEY_TYPE_whiteout)|						\
1505 	  BIT_ULL(KEY_TYPE_hash_whiteout)|					\
1506 	  BIT_ULL(KEY_TYPE_dirent))						\
1507 	x(xattrs,		3,	BTREE_ID_SNAPSHOTS,			\
1508 	  BIT_ULL(KEY_TYPE_whiteout)|						\
1509 	  BIT_ULL(KEY_TYPE_cookie)|						\
1510 	  BIT_ULL(KEY_TYPE_hash_whiteout)|					\
1511 	  BIT_ULL(KEY_TYPE_xattr))						\
1512 	x(alloc,		4,	0,					\
1513 	  BIT_ULL(KEY_TYPE_alloc)|						\
1514 	  BIT_ULL(KEY_TYPE_alloc_v2)|						\
1515 	  BIT_ULL(KEY_TYPE_alloc_v3)|						\
1516 	  BIT_ULL(KEY_TYPE_alloc_v4))						\
1517 	x(quotas,		5,	0,					\
1518 	  BIT_ULL(KEY_TYPE_quota))						\
1519 	x(stripes,		6,	0,					\
1520 	  BIT_ULL(KEY_TYPE_stripe))						\
1521 	x(reflink,		7,	BTREE_ID_EXTENTS|BTREE_ID_DATA,		\
1522 	  BIT_ULL(KEY_TYPE_reflink_v)|						\
1523 	  BIT_ULL(KEY_TYPE_indirect_inline_data)|				\
1524 	  BIT_ULL(KEY_TYPE_error))						\
1525 	x(subvolumes,		8,	0,					\
1526 	  BIT_ULL(KEY_TYPE_subvolume))						\
1527 	x(snapshots,		9,	0,					\
1528 	  BIT_ULL(KEY_TYPE_snapshot))						\
1529 	x(lru,			10,	0,					\
1530 	  BIT_ULL(KEY_TYPE_set))						\
1531 	x(freespace,		11,	BTREE_ID_EXTENTS,			\
1532 	  BIT_ULL(KEY_TYPE_set))						\
1533 	x(need_discard,		12,	0,					\
1534 	  BIT_ULL(KEY_TYPE_set))						\
1535 	x(backpointers,		13,	0,					\
1536 	  BIT_ULL(KEY_TYPE_backpointer))					\
1537 	x(bucket_gens,		14,	0,					\
1538 	  BIT_ULL(KEY_TYPE_bucket_gens))					\
1539 	x(snapshot_trees,	15,	0,					\
1540 	  BIT_ULL(KEY_TYPE_snapshot_tree))					\
1541 	x(deleted_inodes,	16,	BTREE_ID_SNAPSHOT_FIELD,		\
1542 	  BIT_ULL(KEY_TYPE_set))						\
1543 	x(logged_ops,		17,	0,					\
1544 	  BIT_ULL(KEY_TYPE_logged_op_truncate)|					\
1545 	  BIT_ULL(KEY_TYPE_logged_op_finsert))					\
1546 	x(rebalance_work,	18,	BTREE_ID_SNAPSHOT_FIELD,		\
1547 	  BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie))			\
1548 	x(subvolume_children,	19,	0,					\
1549 	  BIT_ULL(KEY_TYPE_set))
1550 
1551 enum btree_id {
1552 #define x(name, nr, ...) BTREE_ID_##name = nr,
1553 	BCH_BTREE_IDS()
1554 #undef x
1555 	BTREE_ID_NR
1556 };
1557 
1558 /*
1559  * Maximum number of btrees that we will _ever_ have under the current scheme,
1560  * where we refer to them with bitfields
1561  */
1562 #define BTREE_ID_NR_MAX		64
1563 
1564 static inline bool btree_id_is_alloc(enum btree_id id)
1565 {
1566 	switch (id) {
1567 	case BTREE_ID_alloc:
1568 	case BTREE_ID_backpointers:
1569 	case BTREE_ID_need_discard:
1570 	case BTREE_ID_freespace:
1571 	case BTREE_ID_bucket_gens:
1572 		return true;
1573 	default:
1574 		return false;
1575 	}
1576 }
1577 
1578 #define BTREE_MAX_DEPTH		4U
1579 
1580 /* Btree nodes */
1581 
1582 /*
1583  * Btree nodes
1584  *
1585  * On disk a btree node is a list/log of these; within each set the keys are
1586  * sorted
1587  */
1588 struct bset {
1589 	__le64			seq;
1590 
1591 	/*
1592 	 * Highest journal entry this bset contains keys for.
1593 	 * If on recovery we don't see that journal entry, this bset is ignored:
1594 	 * this allows us to preserve the order of all index updates after a
1595 	 * crash, since the journal records a total order of all index updates
1596 	 * and anything that didn't make it to the journal doesn't get used.
1597 	 */
1598 	__le64			journal_seq;
1599 
1600 	__le32			flags;
1601 	__le16			version;
1602 	__le16			u64s; /* count of d[] in u64s */
1603 
1604 	struct bkey_packed	start[0];
1605 	__u64			_data[];
1606 } __packed __aligned(8);
1607 
1608 LE32_BITMASK(BSET_CSUM_TYPE,	struct bset, flags, 0, 4);
1609 
1610 LE32_BITMASK(BSET_BIG_ENDIAN,	struct bset, flags, 4, 5);
1611 LE32_BITMASK(BSET_SEPARATE_WHITEOUTS,
1612 				struct bset, flags, 5, 6);
1613 
1614 /* Sector offset within the btree node: */
1615 LE32_BITMASK(BSET_OFFSET,	struct bset, flags, 16, 32);
1616 
1617 struct btree_node {
1618 	struct bch_csum		csum;
1619 	__le64			magic;
1620 
1621 	/* this flags field is encrypted, unlike bset->flags: */
1622 	__le64			flags;
1623 
1624 	/* Closed interval: */
1625 	struct bpos		min_key;
1626 	struct bpos		max_key;
1627 	struct bch_extent_ptr	_ptr; /* not used anymore */
1628 	struct bkey_format	format;
1629 
1630 	union {
1631 	struct bset		keys;
1632 	struct {
1633 		__u8		pad[22];
1634 		__le16		u64s;
1635 		__u64		_data[0];
1636 
1637 	};
1638 	};
1639 } __packed __aligned(8);
1640 
1641 LE64_BITMASK(BTREE_NODE_ID_LO,	struct btree_node, flags,  0,  4);
1642 LE64_BITMASK(BTREE_NODE_LEVEL,	struct btree_node, flags,  4,  8);
1643 LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
1644 				struct btree_node, flags,  8,  9);
1645 LE64_BITMASK(BTREE_NODE_ID_HI,	struct btree_node, flags,  9, 25);
1646 /* 25-32 unused */
1647 LE64_BITMASK(BTREE_NODE_SEQ,	struct btree_node, flags, 32, 64);
1648 
1649 static inline __u64 BTREE_NODE_ID(struct btree_node *n)
1650 {
1651 	return BTREE_NODE_ID_LO(n) | (BTREE_NODE_ID_HI(n) << 4);
1652 }
1653 
1654 static inline void SET_BTREE_NODE_ID(struct btree_node *n, __u64 v)
1655 {
1656 	SET_BTREE_NODE_ID_LO(n, v);
1657 	SET_BTREE_NODE_ID_HI(n, v >> 4);
1658 }
1659 
1660 struct btree_node_entry {
1661 	struct bch_csum		csum;
1662 
1663 	union {
1664 	struct bset		keys;
1665 	struct {
1666 		__u8		pad[22];
1667 		__le16		u64s;
1668 		__u64		_data[0];
1669 	};
1670 	};
1671 } __packed __aligned(8);
1672 
1673 #endif /* _BCACHEFS_FORMAT_H */
1674